""" This global flag controls whether to assign new tensors to the parameters instead of changing the existing parameters in-place when converting an `nn.Module` using the following methods: 1. `module.cuda()` / `.cpu()` (for moving `module` between devices) 2. `module.float()` / `.double()` / `.half()` (for converting `module` to a different dtype) 3. `module.to()` / `.type()` (for changing `module`'s device or dtype) 4. `module._apply(fn)` (for generic functions applied to `module`) Default: False """ _overwrite_module_params_on_conversion = False def set_overwrite_module_params_on_conversion(value): global _overwrite_module_params_on_conversion _overwrite_module_params_on_conversion = value def get_overwrite_module_params_on_conversion(): return _overwrite_module_params_on_conversion